From: Neo Zhang <redacted>
Date: Sat, 29 Nov 2025 12:59:44 +0000 (+0800)
Subject: sycl : support to malloc memory on device more than 4GB, update the doc and script... 
X-Git-Tag: upstream/1.8.3~216
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=a3459484bfc9f5cce61ef6488e1f95de46e5a344;p=pkg%2Fggml%2Fsources%2Fwhisper.cpp

sycl : support to malloc memory on device more than 4GB, update the doc and script (llama/17566)

Co-authored-by: Neo Zhang Jianyu <redacted>
---

diff --git a/ggml/src/ggml-sycl/CMakeLists.txt b/ggml/src/ggml-sycl/CMakeLists.txt
index efd78b91..88f29221 100644
--- a/ggml/src/ggml-sycl/CMakeLists.txt
+++ b/ggml/src/ggml-sycl/CMakeLists.txt
@@ -91,7 +91,10 @@ if (GGML_SYCL_F16)
     add_compile_definitions(GGML_SYCL_F16)
 endif()
 
-if (GGML_SYCL_TARGET STREQUAL "NVIDIA")
+if (GGML_SYCL_TARGET STREQUAL "INTEL")
+    add_compile_definitions(GGML_SYCL_WARP_SIZE=16)
+    target_link_options(ggml-sycl PRIVATE  -Xs   -ze-intel-greater-than-4GB-buffer-required)
+elseif (GGML_SYCL_TARGET STREQUAL "NVIDIA")
     add_compile_definitions(GGML_SYCL_WARP_SIZE=32)
 elseif (GGML_SYCL_TARGET STREQUAL "AMD")
     # INFO: Allowed Sub_group_sizes are not consistent through all
@@ -100,7 +103,8 @@ elseif (GGML_SYCL_TARGET STREQUAL "AMD")
     # Target archs tested working: gfx1030, gfx1031, (Only tested sub_group_size = 32)
     add_compile_definitions(GGML_SYCL_WARP_SIZE=32)
 else()
-    add_compile_definitions(GGML_SYCL_WARP_SIZE=16)
+    # default for other target
+    add_compile_definitions(GGML_SYCL_WARP_SIZE=32)
 endif()
 
 if (GGML_SYCL_GRAPH)
diff --git a/ggml/src/ggml-sycl/cpy.cpp b/ggml/src/ggml-sycl/cpy.cpp
index 1ec99b0a..96709554 100644
--- a/ggml/src/ggml-sycl/cpy.cpp
+++ b/ggml/src/ggml-sycl/cpy.cpp
@@ -515,9 +515,6 @@ void ggml_sycl_cpy(ggml_backend_sycl_context & ctx, const ggml_tensor * src0, co
     const int64_t ne = ggml_nelements(src0);
     GGML_ASSERT(ne == ggml_nelements(src1));
 
-    GGML_ASSERT(ggml_nbytes(src0) <= INT_MAX);
-    GGML_ASSERT(ggml_nbytes(src1) <= INT_MAX);
-
     GGML_TENSOR_BINARY_OP_LOCALS01;
 
     SYCL_CHECK(ggml_sycl_set_device(ctx.device));